Question about insn lengths

[DJ Delorie]

> Note that if you want to do this (with the FSF tools, at least!)
> you'll need to give GCC's branch limits a big kick in the pants
> first.  The MIPS assembler and linker do not know how to lengthen
> short branches.  That's why I've dealt with MIPS errata in GCC in
> the past.

Perhaps the insn that codes the end of loop can have a length
depending on what gcc thinks the length of the loop is, but emit a
pseudo to tell gas to pad according to what gas thinks.  That way, gcc
would never see the loop as smaller than reality because of padding.